Points of novelty

[donaldRwilliams]

Gathering my thoughts together as to what is new and "exciting" about our study (feel free to make additions):

1) First multivariate analysis

• Allow for comparisons between symptoms
  • Important because current medication does not help with negative symptoms and oxytocin is thought of as potentially "better" for negative symptoms

2) First Bayesian analysis

• Continuous measure of evidence
  • probability of, rather than a simple yes/no decision
• Hierarchical shrinkage through which precision is gained
  • Our estimates thus far have narrower intervals than previous studies (about half as wide)

3) First to use standardized mean change (SMCR) from baseline scores

• Selective reporting or simply forking paths in previous meta-analyses
  • A total score (BPRS) for within subjects was substantially higher post-treatment. The between subject score was reported and was lower than the placebo group (Davis)

4) What you have found out about Modabbernia

• Important because each of the previous meta-analyses have not mentioned this.

5) From the analyses thus far, it appears that we will also be the first to report that there is not an effect for any of the symptoms types.

• This is interesting, but also means we have to be careful in what we do and say.

6) First to look at cognitive deficits

• Oxytocin mediates social behavior and social cognition in non-human animals.
  • Evidence is supposedly strong for an effect in healthy humans and clinical populations, but research on healthy subjects has recently been called into question
• Previous meta analyses that were novel were published in high impact journals (eg. Nature)

[paul-buerkner]

I agree with all of them. A few minor comments

(2) It is interesting that our intervals are much smaller than those of the previous MAs. There can also be some other aspects despite hierchical shrinkage contributing to this:

• Moddabernia is currently missing which caused a substantial amount of heterogeneity leading to larger CIs in previous MAs.
• I currently apply an informative prior on tau (between study standard deviation) in form of a half normal prior HN(0, 0.25). This is required in my opinion because 7 or 8 studies are not enough to reasonable estimate tau. If I only used a very weakly informative prior, tau would be unrealistically large (simply because we don't have enough evidence in the data).

(3) We have to make clear, that the SMCR as we currently use it contrasts pre-post changes in experimental and control group. Thus, it incroporates both within and between group differences.

[donaldRwilliams]

It sounds like with SMCR we are asking the right question. What we want to know, for instance, is whether the mean change from baseline is "significantly" different between groups.